It's the startup and REPL time, typically. The JVM is super-powerful when it comes to compute tasks, but define a function on the REPL and Clojure will compile it, which is necessarily slower than what Python does.

So yeah, Clojure's faster, but perception matters. Erlang's a different case because it optimises for responsiveness for small tasks. It's very hard to write Clojure or even raw Java code as good at that as an easy Erlang program.

But yeah, a Mandelbrot set will be faster in Clojure.

JVM startup time is typically sub-second. Anything above that is not the JVM, it's something else.

I agree with you, it's the startup time. It is felt when running unit test too.

That's why JRuby is still a niche and most developers keep using MRI despite JRuby being eventually faster. It's a different example of developer happiness.